CN102662764A - Dynamic cloud computing resource optimization allocation method based on semi-Markov decision process (SMDP) - Google Patents

Abstract

The invention discloses a dynamic cloud computing resource optimization allocation method based on a semi-Markov decision process (SMDP) and belongs to technical field of computer communication. The dynamic cloud computing resource optimization allocation method includes that 1) a could computing service domain system enables user satisfaction to be divided into N classes; 2) a terminal user sends service requests to a cloud computing service domain and applies for use of cloud computing service; 3) the cloud computing service domain system establishes an action set according to the received service requests and current cloud computing service domain states; 4) long-term income of the cloud computing service domain is computed aiming at each action in the action set; and 5) the cloud computing service domain system determines whether to accept the current service requests according to the computed long-term income, if the service requests are accepted, the cloud computing service domain system selects a virtual memory (VM) resource allocation plan corresponding to the action with maximum long-term income to allocate the VM for the cloud computing service requests. Compared with the prior art, the dynamic cloud computing resource optimization allocation method based on the SMDP greatly improves user satisfaction and service quality.

Description

A kind of dynamic cloud computational resource optimized distribution method based on SMDP

Technical field

The invention belongs to the computer communication technology field, relate to the resource optimal distribution method of cloud computing system, relate in particular to the method for distributing rationally that moves in the cloud computing system the cloud computing resource of cloud computing service-domain.

Background technology

Cloud computing is a kind of with resource distribution according to need, pay-as-you-go, and usefulness is calculated as new calculation services pattern (Armbrust, M., Fox, the A. of characteristic; Griffith, R., Joseph, A., Katz; R., Konwinski, A., Lee, G.; Patterson, D., Rabkin, A., Stoica; I., et al. " Above the clouds:A berkeley view of cloud computing " .EECS Department, University of California, Berkeley, Tech.Rep.UCB/EECS-2009-28 (2009)).Cloud computing is not merely cloud computing service provider and for the personal user a kind of new computation schema is provided simultaneously yet; It can be by broad sense be divided into infrastructure-as-a-service (IaaS), platform-as-a-service (PaaS) and software-as-a-service (SaaS) three major types.Along with the development of wireless communication technology and Internet technology, portable terminal will replace PC gradually becomes global topmost internet access facility.Because (movability is for example compared and had more superiority to portable terminal (MD) with traditional catv terminal; Dirigibility and perception etc.); Therefore no matter mobile computing and the cloud computing technology nature that combines is just become and makes up the new method that moves application, still be that industry member has also attracted increasing concern in academia at present.Thereby one new research field-moving cloud computing (Mobile Cloud Computing) also just arises at the historic moment.

Former about in the research of moving cloud computing, what main research direction concentrated on calculation task uploads download, long-range operation and dynamic organization etc.The author is at (X.Li; H.Z; And Y.Zhang, " Deploying Mobile Computation in Cloud Service " in Proceedings of the First International Conference for Cloud Computing (CloudCom), 2009; P.301.) proposed a mobile cloud computing model that can both move mobile application in portable terminal and high in the clouds in, thereby the portable terminal of resource-constrained can upload to the high in the clouds operation with calculating, transmission and store tasks.The author is at (B.Chun and P.Maniatis; " Augmented Smartphone Applications Through Clone Cloud Execution; " In Proceedings of USENIX HotOS XII; 2009.) in dispose CloneCloud cloud resource through increase carrying out number of times, but do not consider the actual motion state of user terminal.Portable terminal through system for cloud computing to the resources allocation of flexible application service at (X.Zhang, J.Schiffman, S.Gibbs; A.Kunjithapatham; And S.Jeong, " Securing elastic applications on mobile devices for cloud computing, " in Proceedings of the 2009 ACM workshop on Cloud computing security; 2009, pp.127-134.) some preliminary researchs have been done in the inside.At document (D.Huang; X.Zhang; M.Kang, and J.Luo, " Mobicloud:A secure mobile cloud framework for pervasive mobile computing and communication; " In Proceedings of 5th IEEE International Symposium on Service-Oriented System Engineering; 2010.) lining, people such as Huang have proposed mobile cloud computing framework, and this model allows portable terminal related application to be uploaded to virtual machine (VM-Virtual Machine) operation in high in the clouds.The author is at (X.Meng, V.Pappas, and L.Zhang; " Improving the scalability of data center networks with trafficaware virtual machine placement, " in IEEE INFOCOM, San Diego; CA; USA, March 2010.) in proposed a kind of different flow and come configuring virtual machine according to different regions, improve the new method of the utilization factor of network through the placement location of distributing virtual machine rationally.In fact, because these research and inquirement about the framework facility of mobile cloud computing are relatively more abundant, therefore, the resources allocation of moving cloud computing will become Next main direction of studying naturally.

In moving system for cloud computing, based on the distributed placement of server group on the geographic position, the cloud computing resource of system (for example CPU, internal memory and storage etc.) is responsible for distribution by a plurality of mobile cloud computing service-domains respectively.Each moves the cloud computing service-domain is made up of a plurality of virtual machines (VM-Virtual Machine), and each virtual machine (VM) then is made up of the minimum cloud computing resource that can handle a cloud computing service.Although compare with portable terminal, the cloud computing resource that moves system for cloud computing is considered to unlimited usually, still is necessary very much to make full use of the low cost movement that the cloud computing resource that moves in the cloud computing service-domain realizes moving system for cloud computing.

Resource optimization to the especially mobile cloud computing of cloud computing distributes the research of carrying out also fewer at present.Document (H.Liang; D.Huang; And D.Peng; " On Economic Mobile Cloud Computing Model, " in Proceedings of the International Workshop on Mobile Computing and Clouds (MobiCloud in conjunction with MobiCASE), 2010.) an Eco-power mobile cloud computing resource allocator model has been proposed; This model can be under the situation of given system configuration, moves the maximum return that should be used for obtaining to move system for cloud computing through optimized distribution beyond the clouds and between the portable terminal.Document (G.Wei; A.V.Vasilakos; Y.Zheng, and N.Xiong, " A game-theoretic method of fair resource allocation for cloud computing services; " 2009.) having proposed one based on game theoretic cloud computing resource allocator model, this model can distribute the cloud computing resource to the demand of QoS of customer (QoS) according to portable terminal.In addition, also have some documents how to come optimized distribution cloud computing resource to study through the server of virtual machine or data center to system for cloud computing.At (K.Lorincz, B.r.Chen, J.Waterman; G.Werner-Allen, and M.Welsh, " Resource aware programming in the pixie os; " In SenSys ' 08, Raleigh, North Carolina; USA, November 2008.) in, the author has proposed a new cloud computing operation model; This operation model can not only make the user under the situation of grasping the cloud computing resource, programme, and can realize that also the allocation model of cloud computing resource is reused in the cloud computing service in the system for cloud computing simultaneously.Document (K.Lorincz, B.Chen, J.Waterman; G.Werner-Allen, and M.Welsh, " A stratified approach for supporting high throughput event processing applications; " In DEB S ' 09, Nashville, TN; USA July2009.) studies the cloud computing resources allocation of event application in the system for cloud computing.At (G.Tesauro, N.K.Jong, R.Das; And M.N.Bennani, " A hybrid reinforcement learning approach to autonomic resource allocation, " in Proc.of ICAC-06; Dublin, Ireland, June 2006.) in; The author has proposed a resource allocator model based on the enhancement mode self learning system and has come the server in the system for cloud computing is carried out dynamic assignment, thereby improves the income of system for cloud computing.At (K.Boloor; R.Chirkova; Y.Viniotis; And T.Salo, " Dynamic request allocation and scheduling for context aware applications subject to a percentile response time sla in a distributed cloud, " in 2 ^NdIEEE International Conference on Cloud Computing Technology and Science; Indianapolis, Indiana, USA; November 2010.) in; The author has proposed a general scheme that the cloud computing services request is distributed and planned, this scheme has improved cloud computing service provider's income when obtaining user's service specified quality.

To the optimized distribution of cloud computing resource, domestic some solutions that also proposed.Such as in patented claim 201110097395.8 (a kind of management control system for cloud computing technological system), author (Cao Xuezhu) has proposed a kind of invention of managing control system for cloud computing technological system; In patented claim 201110138021.6 (a kind of cloud computing resource management system and method), author's (Du Hai and horse are strong for Ji Xinhua, Nie Song) has proposed the invention of a kind of cloud computing resource management system and method; In patented claim 201110075410.9 (management method of configuration information and system in the cloud computing operating system), author's (Zhang Liqiang and Haitao Zhang) has proposed the management method of configuration information in a kind of cloud computing operating system and the invention of system; In patented claim 201080005003.4 (being used for the system and method in the automatic managing virtual resource of cloud computing environment), author (SM You Mubaihaoke) has proposed a kind of invention that is used in the system of cloud computing environment managing virtual resource; In patented claim 201110222073.1 (a kind of cloud computing management system based on virtual resource), author (Shen Lingyun, Ruan Minhui and Zhou Yongfeng) has proposed a kind of invention of the cloud computing management system (C2MS) based on virtual resource.A main advantage that moves system for cloud computing is to allow portable terminal to move their mobile application service beyond the clouds.And the cloud computing resource that cloud computing service can also be assigned with a plurality of VM makes portable terminal obtain higher calculating and storage capacity.Receive one during from cloud computing services request that portable terminal sends over when moving the cloud computing service-domain, system need analyze current available cloud computing resource, and whether decision receives this cloud computing services request based on analysis result; If decision is to receive, system also needs further to adjudicate the cloud computing services request that is specially this portable terminal and distributes how many cloud computing resources (being the number of VM) so.If it is occupied to move cloud computing resources all in the cloud computing service-domain, owing to the deficiency of cloud computing resource, the cloud computing services request (we suppose in moving cloud computing, do not have buffer queue) of this portable terminal can be refused by system so.Refusal to portable terminal cloud computing services request has not only brought negative influence to mobile terminal user satisfaction and service quality, and has greatly reduced the net proceeds of system.

The system's income that moves the cloud computing service-domain increases along with the increase of received cloud computing services request quantity usually.But then; The cloud computing services request that receives along with system is many more; The cloud computing resource of distributing to each cloud computing service so is also just few more, thereby has reduced the mobile terminal user satisfaction accepting to serve and the system performance of mobile cloud computing service-domain.And have the income of only having considered system about cloud computing resource allocation methods major part now, and do not consider the occupied expenditure of bringing of cloud computing resource, do not consider mobile terminal user satisfaction and service quality (QoS) yet.Therefore; In order to obtain moving the comprehensive system benefit of cloud computing service-domain; When calculating the system benefit that moves the cloud computing service-domain; Not only need consider to move the income of system for cloud computing, also need consider occupied expenditure of bringing of cloud computing resource and mobile terminal user satisfaction and service quality (QoS).

Summary of the invention

Optimized distribution problem to the cloud computing resource that moves system for cloud computing cloud computing service-domain the object of the present invention is to provide a kind of dynamic cloud computational resource optimization method based on SMDP.The present invention has newly proposed the mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model based on half Ma Shi decision process (SMDP); Obtain the optimized distribution decision policy of the cloud computing resource of mobile cloud computing service-domain through this model; And obtain moving the maximum return of cloud computing service-domain; This income has not only considered to receive the income that the cloud computing services request is brought; Also considered because of the cloud computing service takies the expenditure that the cloud computing resource is brought simultaneously, and mobile terminal user satisfaction and service quality (QoS).Therefore, this invention all has important effect to the integral benefit that moves cloud computing system and portable terminal client to the raising of the satisfaction that moves system for cloud computing, and this also is actual value of the present invention place.

Technical scheme of the present invention is:

A kind of dynamic cloud computational resource optimized distribution method based on SMDP the steps include:

1) cloud computing service-domain system is divided into the N class with user satisfaction, and the satisfaction classification is that the corresponding virtual machine VM number that distributes of the user of i is k _iWherein, 1≤k _i≤K, K are the VM sum in the cloud computing service-domain;

2) terminal user sends services request and gives the cloud computing service-domain, and the cloud computing service is used in application;

3) cloud computing service-domain system sets up an action set according to services request that receives and current cloud computing service-domain state;

4), calculate the long-term gain of cloud computing service-domain to each action in the said action set;

5) cloud computing service-domain system determines whether to accept the current service request according to the long-term gain of calculating, and is cloud computing services request distribution VM if accept then choose the corresponding VM Resource Allocation Formula of the maximum action of long-term gain.

Further, the state s of cloud computing service-domain is expressed as s=<n ₁, n ₂..., n _N, e>Wherein, n _iFor satisfaction classification in the cloud computing service-domain is the number of users of i, e is the incident in the cloud computing service-domain, e ∈ { R, D ₁, D ₂, D _i...., D _N, R is the cloud computing services request, D _iFor the satisfaction classification is the cloud computing service completion of i and has discharged its shared VM number.

Further, said action set does $A\left(s\right)=\left\{\begin{array}{cc}-1& e\&Element;\{{\mathrm{<figure-callout\; id="1"\; label="D"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">D</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">D</figure-callout>}}_2,...,D_N\}\\ \{\mathrm{0,1},...,N\},& e=R.\end{array}\right.;$ Wherein, A (s)=-1 expression cloud computing service finishes operation and discharges shared cloud computing state of resources; A (s)=0 expression cloud computing service-domain refusal cloud computing services request, A (s)=i representes that the cloud computing resource that the cloud computing service-domain receives the cloud computing services request and distributes to this cloud computing services request is k _iIndividual VM; S representes the current state of cloud computing service-domain.

Further, utilize formula z (s, a)=x (s, a)-τ (s, a) y (s, a) calculate income z to each action a (s, a); Wherein, and x (s is that state is the action of s, selection when being a a), the gross income that the cloud computing service-domain is obtained, and (s when the action that a) to be illustrated in state be s, choose is a, transfers to desired service time of next state j to τ; Y (s, the expenditure of cloud computing service-domain when the action that a) to be illustrated in state be s, choose is a.

Further, utilize formula $x(s,a)=\left\{\begin{array}{cc}-1,& e=R,a=0\\ U\left(k_i\right),& e=R,a=i\end{array}\right.$ Calculate the gross income that the cloud computing service-domain is obtained, U (k _i) be the usefulness function.

Further, utilize formula

Calculate the cloud computing service-domain and transfer to next state desired service time by current state; Wherein, α is the discount rate under continuous time between two decision points, and decision point is meant the time point that any one incident e takes place, τ ₁Be meant the time that state experienced from current state to next incident generation.

Further, and the time τ between two decision points (s, a) obeys index distribution, the mean speed γ that incident takes place (s, a)=τ (s, a) ^-1

Further, cloud computing service-domain system utilizes formula $v\left(s\right)=z(s,a)+\mathrm{\&eta;}\underset{j\&Element;S}{\mathrm{\&Sigma;}}p\left(j\right|s,a)v\left(j\right),$ The long-term gain v (s) of cloud computing service-domain when calculating each the action a in the said action set; Wherein,

(j|s a) is state transition probability to p, and j is for moving the NextState of cloud computing service-domain; γ (s; A) mean speed that takes place for incident, α is the discount rate under continuous time between two decision points, decision point is meant the time point that any one incident e takes place; S is all possible states of cloud computing service-domain, the whole long-term gain that is obtained during v (j) expression NextState j.

Further, adopt a usefulness function to measure cloud computing user's satisfaction, user satisfaction is divided into the N class.

Compared with prior art, good effect of the present invention is:

The present invention is based on half Ma Shi decision process (SMDP); A new mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model has at first been proposed; The dynamic cloud computational resource optimized distribution decision policy that obtains through this model can not only make the system benefit that moves the cloud computing service-domain maximum, also can improve the utilization factor and mobile terminal user satisfaction and the service quality (QoS) that move cloud computing service-domain cloud computing resource simultaneously.With traditional greedy algorithm the allocative decision of Internet resources is compared, according to the optimisation strategy that the mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model that we proposed obtains, its system benefit and performance have all had significantly raising.Can know by Fig. 5 and Fig. 6; Growth along with cloud computing services request arrival rate; Especially when the arrival rate of cloud computing services request surpasses 5; Efficiency earnings of the present invention is compared with traditional greedy algorithm and has been improved (as shown in Figure 5) more than at least 50%, and blocking rate of the present invention is compared with traditional greedy algorithm and then reduced (as shown in Figure 6) more than 50% at least simultaneously.

The main contribution of the present invention is embodied in following three aspects:

1) derived the dynamic cloud computational resource optimized distribution decision policy of mobile cloud computing service-domain based on half Ma Shi decision process (SMDP).

2) this model can be based on the current available cloud computing resource of mobile cloud computing service-domain; For the cloud computing services request is distributed different cloud computing resources adaptively; Improve the cloud computing resource utilization through making full use of this cloud computing resource that moves the cloud computing service-domain, and obtain to move the largest global income of cloud computing service-domain.

3) the maximum system income of the mobile cloud computing service-domain of this model acquisition; Both considered that this moved the cloud computing service-domain and receives the income that the cloud computing services request is brought; Also considered also to have considered mobile terminal user satisfaction and service quality (QoS) because of the occupied expenditure of bringing of cloud computing resource.Therefore, the system benefit that obtains through this model is comprehensive integral benefit.

Mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model proposed by the invention not only can improve the utilization factor of the cloud computing resource that moves system for cloud computing cloud computing service-domain, also can improve mobile subscriber's service quality (QoS) simultaneously.In order to verify the performance of mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model proposed by the invention, we have carried out comparison (R.Ramjee, D.Towsley through experiment with its performance with traditional greedy algorithm (Greedy Algorithm); And R.Nagarajan, " On optimal call admission control in cellular networks, " Wireless Networks; Vol.3; No.1, pp.29-41,1997).Our experimental result shows; Use the mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model that the present invention proposes; The entire system income of mobile system for cloud computing is compared with greedy algorithm and has been improved more than 50%; The unaccepted probability of its cloud computing services request is compared with greedy algorithm and has then been reduced more than 50%, also is that performance and the service quality (QoS) of mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model proposed by the invention is compared all with greedy algorithm and improved more than 50%.

Description of drawings

Fig. 1 is for moving the service model of system for cloud computing;

Fig. 2 is a method flow diagram of the present invention;

Fig. 3 is the usefulness function of multimedia service;

Fig. 4 is state transition diagram (N=2); Wherein, Connect the action that first (for example a=0) on the arrow line of two states is illustrated under the current state to be taked; Second (for example

) is illustrated under the current state on the arrow line of two states of connection; After taking corresponding action, transfer to the transition probability of next state;

Fig. 5 efficiency earnings of the present invention and traditional greedy algorithm comparison diagram;

Fig. 6 blocking rate of the present invention and traditional greedy algorithm comparison diagram.

Embodiment

Below in conjunction with accompanying drawing the present invention is explained in further detail.

1. move the arthmetic statement of cloud computing service-domain dynamic cloud computational resource optimized distribution model:

Moving the main advantage that system for cloud computing compares with traditional Client-Server service mode is: when portable terminal uploads to high in the clouds when carrying out computing with their application service; Portable terminal can obtain more capacity and more performance (for example still less processing time, the saving of battery capacity of mobile terminal etc.).Uploading of portable terminal flexible application task can realize through the Weblet that connects high in the clouds and portable terminal.Java or a .Net or the Python that Weblet can use to be independent of platform also can the usage platform programming language.At (B.Chun and P.Maniatis; " Augmented Smartphone Applications Through Clone Cloud Execution; " In Proceedings of USENIX HotOS XII, 2009.) studied the algorithm that Weblet is uploaded to the high in the clouds operation from portable terminal in.Upload flexible application through Weblet and serve high in the clouds and move, portable terminal can significantly improve computing power, storage capacity and the network bandwidth etc. of self.Usually, whether the portable terminal decision uploads to the state (for example, electric weight, network connection quality and the portable terminal of the CPU processing power of portable terminal, battery are to the factors such as consideration of safety) that portable terminal self is depended in the high in the clouds operation with task.In the present invention; When the portable terminal decision uploads to the high in the clouds operation with task; It can at first send a services request to high in the clouds, if the services request of portable terminal has been accepted in high in the clouds, portable terminal will upload to the high in the clouds operation with task subsequently so; Behind the end of run, high in the clouds can return to portable terminal with operation result.

In the present invention, the computational resource and the communication resource (comprising the CPU in the server, memory device, internal memory etc., and other routing devices and communication facilities etc.) that move in the system for cloud computing all are to come unified management by virtual machine (VM).As shown in Figure 1, a VM is in charge of Weblet uploading, unloading and handling in moving system for cloud computing.As previously mentioned; In mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model proposed by the invention; A VM handles a required minimum cloud computing resource (CPU of cloud computing service in moving the cloud computing service-domain; Internal memory and storage etc.), the assigned cloud computing resource of each VM once can only be handled a cloud computing services request.Though we can think that the cloud computing resource that moves in the system for cloud computing is unlimited, in moving system for cloud computing, the cloud computing resource that certain concrete mobile cloud computing service-domain is counted with VM quantity is again limited.Therefore, in mobile cloud computing service-domain, if the quantity of the cloud computing services request that arrives has surpassed cloud computing resource VM number available in this service-domain, the cloud computing services request that then arrives subsequently will be by this service-domain refusal.On the other hand; If the quantity of the cloud computing services request that arrives is far below cloud computing resource VM number available in this service-domain; This service-domain just can distribute more VM number for each cloud computing services request and makes full use of the cloud computing resource of this service-domain so, improves this cloud computing resource utilization that moves the cloud computing service-domain and mobile terminal user satisfaction and service quality (QoS) with this.

Therefore; The target of mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model proposed by the invention is exactly through making full use of the cloud computing resource; Make that moving the cloud computing service-domain can obtain the largest global system benefit, also can improve cloud computing resource utilization and the user satisfaction and the service quality (QoS) of this service-domain.

In the present invention, we consider a mobile system for cloud computing that only comprises a cloud computing service-domain, establish its cloud computing resource and are total up to K virtual machine (VM).Represent to distribute to the VM number of the cloud computing resource of a portable terminal with k, wherein k is a positive integer, and the 1≤k that satisfies condition≤K.In addition, we can use different usefulness function (J.W.Lee, R.R.Mazumdar; And N.B.Shroff, " Non-convex optimization and rate control for multi-class services in the Internet, " IEEE/ACM Transactions on Networking; 2005; Vol.13, no.4 pp.827-840.) measures the satisfaction that moves the cloud computing user.For example, can describe with similar Sigmoidal function and move cloud computing mobile terminal user satisfaction,

$U\left(r\right)=1-\exp (-\frac{{\mathrm{\&omega;}}_2{\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">r</figure-callout>}}^2}{{\mathrm{\&omega;}}_1+r}),---\left(1\right)$

Wherein the user satisfaction of system for cloud computing is moved in U (r) expression, and r moves the cloud computing resource that the cloud computing service-domain is distributed to portable terminal, ω ₁And ω ₂Be the parameter that is used for regulating U (r) waveform, the waveform of its function is as shown in Figure 3.

Usually, parameter ω ₁And ω ₂Selection be with the final user demand of service quality (QoS) to be decided by moving the cloud computing service, effectively index selects that the cloud computing resources allocation of mobile cloud computing service-domain is had remarkable influence.Can know from formula (1),, need distribute cloud computing resource as much as possible to mobile phone users in order to improve the user satisfaction that moves system for cloud computing.But; On the other hand; In order to improve the overall system income that moves system for cloud computing; Move the total available cloud computing resource of cloud computing service-domain and the mobile subscriber uses the demand of cloud computing service to the cloud computing resource according to this, system can not all distribute maximum cloud computing resource for separately each mobile phone users again.In order to set up Optimization Model to the dynamic need that the cloud computing of moving system for cloud computing is served; We suppose that mobile terminal request inserts the process obedience Poisson distribution (Poisson) of moving system for cloud computing and using the cloud computing service; Its average is λ; That also supposes mobile cloud computing terminal user simultaneously is netting the time obeys index distribution; Its average representes to reach with Poisson distribution the cloud computing rate request of system for cloud computing for λ, and μ representes that the user who finishes to serve leaves the speed of system for cloud computing obeys index distribution.

2. as shown in Figure 2, it is following to set up the step that moves cloud computing service-domain dynamic cloud computational resource optimized distribution model:

1) system state is set

In order to use half Ma Shi decision process to characterize the optimized distribution model that moves cloud computing service-domain cloud computing resource; Because the user is inversely proportional to satisfaction that moves system for cloud computing and the time that its services request is processed; Be that the shorter then user of time that is processed of user's services request is high more to the satisfaction of system; Obviously, if it is many more to distribute to this user's cloud computing resource (being the VM number), then time of being processed of this user institute requested service is then shorter (as a rule; The cloud computing service can come parallel processing by the cloud computing resource of a plurality of VM, thereby improves travelling speed).Hence one can see that, and the user is directly proportional with the cloud computing resource of distributing to this user to the satisfaction that moves system for cloud computing, and the cloud computing resource of promptly distributing to the user is many more, and then this user's satisfaction is high more.The user satisfaction that we will move system for cloud computing is divided into the N class.Therefore to the present invention is based on the system state of the mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model of half Markovian process be the cloud computing quantity of service that had under each user satisfaction and move the set of institute's event in the cloud computing service-domain at this in our definition.We also define k _iFor distributing to the VM number of cloud computing resource that user satisfaction is the user of i, i=1 here, 2 ..., N, and 0＜k ₁＜...＜k _N≤K.Use n _iUser satisfaction is all numbers of users of i in the mobile cloud computing service-domain of expression.Satisfaction is that the user of i should distribute k _iThe cloud computing resource of individual VM.

In mobile cloud computing service-domain, always have two types incident:

1) a cloud computing services request that newly arrives is represented with R;

2) user satisfaction is i operation has been accomplished in the cloud computing service, and has discharged its shared cloud computing resource, uses D _iRepresent.

Therefore any incident e can use e ∈ { R, D in this moves the cloud computing service-domain ₁, D ₂...., D _NRepresent that all possible states of system are represented with S, thus the system state that moves the cloud computing service-domain can be represented with following formula:

S＝{s|s＝<n ₁，n ₂，...，n _N，e>}.

(2)

2) the action set is set

When a terminal user asks to insert mobile cloud computing service-domain and application use cloud computing service (e=R); The cloud computing service-domain should be moved and this user's request need be determined whether to accept; If accept, the cloud computing resource that should distribute what VM to the user of this request service so.Be simple meter, we should move cloud computing services request of cloud computing service-domain refusal with A (s)=0 expression; Represent to move the cloud computing service-domain with A (s)=i and received this cloud computing services request, and the cloud computing resource of distributing to this cloud computing services request is k _iIndividual VM reaches i in the hope of making the terminal user to the satisfaction of this cloud computing service, and s representes current system state here.And on the other hand, we are illustrated in cloud computing service with A (s)=-1 and finish operation and discharge shared cloud computing state of resources (incident e=D here _i) under action, i.e. the VM number of the existing available cloud computing resource of statistics and wait for the generation of next incident.Therefore, the action of this model set is summed up as follows:

$A\left(s\right)=\left\{\begin{array}{cc}-1& e\&Element;\{{\mathrm{<figure-callout\; id="1"\; label="D"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">D</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="D"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">D</figure-callout>}}_2,...,D_N\}\\ \{\mathrm{0,1},...,N\},& e=R.\end{array}\right.---\left(3\right)$

3) earnings pattern is set

Based on system state and corresponding action; We can estimate an obtainable income of mobile cloud computing service-domain (with z (s; A) represent; A is the action that system takes each incident, comprises refusing the cloud computing services request is perhaps distributed i VM for the cloud computing services request cloud computing resource.Under the arrival state of cloud computing services request, a also is appreciated that the VM number for the cloud computing resource of distributing to each user; But under the state that the mobile subscriber leaves after the cloud computing service finishes, a=-1 representes to add up existing available cloud computing resource.), this income is made up of two parts, and the one, the income of system, another part is the expenditure of system, can represent with following formula,

z(s，a)＝x(s，a)-τ(s，a)y(s，a) (4)

X (s is that system is s at state a), and when the action of selection was a, the gross income that system obtained can be expressed as,

$x(s,a)=\left\{\begin{array}{cc}-1,& e=R,a=0\\ U\left(k_i\right),& e=R,a=i\end{array}\right.---\left(5\right)$

Wherein, U (k _i) be the usefulness function, shown in formula (1).τ (s, a) being illustrated in current system state is s, when the action of choosing is a, transfers to desired service time of next system state j; Y (s, a) being illustrated in current system state is s, the expenditure the when action of choosing is a, y (s a) can measure with total number of the shared cloud computing resource VM of the cloud computing that moving service, be expressed from the next into:

$y(s,a)=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{n}_i{k}_i.---\left(6\right)$

The present invention is based on the long-term expected revenus that earnings pattern obtains and determine whether this request of decision.The long-term gain that long-term gain model of the present invention can be received respectively, refuse according to earnings pattern is selected maximum that action a (receiving or refusal) of long-term gain then; If received current request, can the update system state.Can wait for next incident simultaneously, then in that incident decision action, and update system state again, the operation that goes round and begins again is like this gone down.

4) solve state transition probability

Decision point is meant the time point that takes place when any one incident; For example a cloud computing services request arrives and moves the cloud computing service-domain, or a mobile phone users of having accomplished use cloud computing service leaves this cloud computing service-domain and discharges shared cloud computing resource.In our system model, because the time τ between two decision points (s, a) equal obeys index distribution, therefore, the mean speed γ that all incidents take place (s a) can be expressed as,

$\mathrm{\&gamma;}(s,a)=\mathrm{\&tau;}{(s,a)}^{-1}$

$=\left\{\begin{array}{cc}\mathrm{\&lambda;}+\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{n}_i\mathrm{\&mu;},& e=R,a=0\\ & \mathrm{ore}=D_i\\ \mathrm{\&lambda;}+(\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{n}_i+1)\mathrm{\&mu;},& e=R,a=i\end{array}\right.---\left(7\right)$

Thus, time τ (s, the expectation discount income z between a) (s a) can be expressed as,

$z(s,a)=x(s,a)-y(s,a)E_s^a\left\{{\&Integral;}_0^{{\mathrm{\&tau;}}_1}{\exp }^{-\mathrm{\&alpha;t}}\mathrm{dt}\right\}$

$=x(s,a)-y(s,a)E_s^a\left\{\frac{1-{\exp }^{-{\mathrm{\&alpha;\&tau;}}_1}}{\mathrm{\&alpha;}}\right\}---\left(8\right)$

$=x(s,a)-\frac{y(s,a)}{\mathrm{\&alpha;}+\mathrm{\&gamma;}(s,a)}$

Wherein, E is under state s, when taking action a, to the average expected time that next incident takes place, τ ₁Be meant the time that state experienced from current state to next incident generation.X (s; A) and y (s; A) define at formula (5) and (6) respectively, owing to the time between two incidents is continuous, and we disperse at the time point of doing decision-making; Need be through the income of continuous time be changed the income that just can obtain discrete time, so we represent the discount rate under continuous time with α.We let p, and (j|s, a) expression system be at state s, and when the action of choosing is a, system transfers to the transition probability of state j.We can derive all state transition probabilities thus.Write in order to simplify, the symbol that we are defined as follows,

${\hat{n}}_1=<{\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">n</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">n</figure-callout>}}_2,..,n_i,..,n_N>$

${\hat{n}}_{2,i}=<{\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">n</figure-callout>}}_1,..,n_i-1,..,n_N>$

${\hat{n}}_{3,i}=<{\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">n</figure-callout>}}_1,..,n_i+1,..,n_N>$

${\hat{n}}_{4,i,m}=<{\mathrm{<figure-callout\; id="1"\; label="n"\; filenames="HDA0000157323010000021.png,HDA0000157323010000031.png"\; state="\{\{state\}\}">n</figure-callout>}}_1,..,n_i+1,..,n_m-1,..,n_N>.---\left(9\right)$

When a new service request arrives at the mobile cloud computing cloud computing service domain, then the system if the decision is rejected, then there is while a = 0; or when a user satisfaction is i run end cloud computing services, the mobile end-user to leave this mobile cloud computing cloud computing service domain and release the resource, then the system which is receiving cloud services to reduce the number of users and available cloud computing resources increases, so there is In both cases, we can get the transition probability as,

$p\left(j\right|s,a)=\left\{\begin{array}{cc}\frac{\mathrm{\&lambda;}}{\mathrm{\&gamma;}(s,a)},& j=<{\hat{n}}_1,R>\\ \frac{n_{i,\mathrm{\&mu;}}}{\mathrm{\&gamma;}(s,a)},& j=<{\hat{n}}_{2,i},D_i>,n_i\&GreaterEqual;1.\end{array}\right.---\left(10\right)$

When the cloud computing service-domain is moved in a new cloud computing services request arrival, if system's decision-making at this moment is that agreement access and the cloud computing resource that is prepared as this cloud computing services request distribution are k _iIndividual VM, this moment so

A=i is arranged simultaneously, i=1,2 ..., N, in this case, we can obtain transition probability and do,

$p\left(j\right|s,a)=\left\{\begin{array}{cc}\frac{(n_i+1)\mathrm{\&mu;}}{\mathrm{\&gamma;}(s,a)},& j=<{\hat{n}}_1,D_i>\\ \frac{\mathrm{\&lambda;}}{\mathrm{\&gamma;}(s,a)},& j=<{\hat{n}}_{3,i},R>\\ \frac{n_{m,\mathrm{\&mu;}}}{\mathrm{\&gamma;}(s,a)},& j=<{\hat{n}}_{4,i,m},D_m>,n_m\&GreaterEqual;1,m\&NotEqual;i.\end{array}\right.---\left(11\right)$

Fig. 4 has provided the mobile system for cloud computing dynamic cloud computational resource optimized distribution decision model that proposes based on us, the state transition diagram during N=2.

5) solve maximized entire system long-term gain

Thus, according to the definition (SMDP) of half Ma Shi decision process, the maximum long-term discount income that we can obtain mobile cloud computing service-domain dynamic cloud computational resource optimized distribution model based on SMDP proposed by the invention does,

$v\left(s\right)=\underset{a\&Element;A\left(s\right)}{\max }\{z(s,a)+\mathrm{\&eta;}\underset{j\&Element;S}{\mathrm{\&Sigma;}}p\left(j\right|s,a)v\left(j\right)\}---\left(12\right)$

z (s wherein; A) and p (j|s; A) respectively in formula (8); (10) Yu in (11) obtain the whole long-term gain that v (j) expression system is obtained when next state j.

6) find optimized decision-making

According to the resulting maximum system income of formula (12); We can find and the corresponding system decision-making of this maximum return easily; This decision-making is the optimal decision-making that current mobile cloud computing service-domain dynamic cloud computational resource is optimized distribution, is the optimal decision-making strategy that moves cloud computing service-domain dynamic cloud computational resource optimized distribution model and move the strategy that the optimal decision-making of cloud computing service-domain dynamic cloud computational resource optimized distribution forms by all.

According to step 5) incident is taked to obtain corresponding long-term gain respectively after each action; Then in these long-term gain the insides; Select the maximum corresponding action of that long-term gain (promptly refuse the cloud computing services request, perhaps distribute the VM number of cloud computing resource) for the cloud computing services request.

Claims (9)

1. the dynamic cloud computational resource optimized distribution method based on SMDP the steps include:

1) cloud computing service-domain system is divided into the N class with user satisfaction, and the satisfaction classification is that the corresponding virtual machine VM number that distributes of the user of i is k _iWherein, 1≤k _i≤K, K are the VM sum in the cloud computing service-domain;

2) terminal user sends services request and gives the cloud computing service-domain, and the cloud computing service is used in application;

3) cloud computing service-domain system sets up an action set according to services request that receives and current cloud computing service-domain state;

4), calculate the long-term gain of cloud computing service-domain to each action in the said action set;

5) cloud computing service-domain system determines whether to accept the current service request according to the long-term gain of calculating, and is cloud computing services request distribution VM if accept then choose the corresponding VM Resource Allocation Formula of the maximum action of long-term gain.

2. the method for claim 1 is characterized in that the state s of cloud computing service-domain is expressed as s=<n ₁, n ₂..., n _N, e>Wherein, n _iFor satisfaction classification in the cloud computing service-domain is the number of users of i, e is the incident in the cloud computing service-domain, e ∈ { R, D ₁, D ₂, D _i...., D _N, R is the cloud computing services request, D _iFor the satisfaction classification is the cloud computing service completion of i and has discharged its shared VM number.

3. method as claimed in claim 2 is characterized in that said action set does $A\left(s\right)=\left\{\begin{array}{cc}-1& e\&Element;\{D_1,D_2,...,D_N\}\\ \{\mathrm{0,1},...,N\},& e=R.\end{array}\right.;$ Wherein, A (s)=-1 expression cloud computing service finishes operation and discharges shared cloud computing state of resources; A (s)=0 expression cloud computing service-domain refusal cloud computing services request, A (s)=i representes that the cloud computing resource that the cloud computing service-domain receives the cloud computing services request and distributes to this cloud computing services request is k _iIndividual VM; S representes the current state of cloud computing service-domain.

4. method as claimed in claim 3, it is characterized in that utilizing formula z (s, a)=x (s, a)-τ (s, a) y (s, a) calculate income z to each action a (s, a); Wherein, and x (s is that state is the action of s, selection when being a a), the gross income that the cloud computing service-domain is obtained, and (s when the action that a) to be illustrated in state be s, choose is a, transfers to desired service time of next state j to τ; Y (s, the expenditure of cloud computing service-domain when the action that a) to be illustrated in state be s, choose is a.

5. method as claimed in claim 4 is characterized in that utilizing formula $x(s,a)=\left\{\begin{array}{cc}-1,& e=R,a=0\\ U\left(k_i\right),& e=R,a=i\end{array}\right.$ Calculate the gross income that the cloud computing service-domain is obtained, U (k _i) be the usefulness function.

6. method as claimed in claim 4 is characterized in that utilizing formula

Calculate the cloud computing service-domain and transfer to next state desired service time by current state; Wherein, α is the discount rate under continuous time between two decision points, and decision point is meant the time point that any one incident e takes place, τ ₁Be meant the time that state experienced from current state to next incident generation.

7. method as claimed in claim 6, it is characterized in that two between the decision point time τ (s, the mean speed γ that a) obeys index distribution, incident take place (s, a)=τ (s, a) ^-1

8. like the arbitrary described method of claim 3～7, it is characterized in that cloud computing service-domain system utilizes formula $v\left(s\right)=z(s,a)+\mathrm{\&eta;}\underset{j\&Element;S}{\mathrm{\&Sigma;}}p\left(j\right|s,a)v\left(j\right),$ The long-term gain v (s) of cloud computing service-domain when calculating each the action a in the said action set; Wherein,

(j|s a) is state transition probability to p, and j is for moving the NextState of cloud computing service-domain; γ (s; A) mean speed that takes place for incident, α is the discount rate under continuous time between two decision points, decision point is meant the time point that any one incident e takes place; S is all possible states of cloud computing service-domain, the whole long-term gain that is obtained during v (j) expression NextState j.

9. the method for claim 1 is characterized in that adopting a usefulness function to measure cloud computing user's satisfaction, and user satisfaction is divided into the N class.

Images